Data Science Solutions

We are data scientists community aspired to leverage data analysis and machine learning expertise appling to business challenges

Our work samples

Industry: medicine

Detection and clustering of unicellular organism cells

The pipeline consists of several steps:
Technologies: yolov5, pythorch, sklearn, plotly, tsne.
Industry: retail

AI assistant of retail articles search.

System triggers to make a picture of items, which customer puts on the scale, and predicts labels of top probable classes. The pretrained convolutional neural network model has been used for transfer learning. Model is deployed in the cloud and accessible through API. Converted model is also available to run in the offline points of sale. There are several classifiers deployed in production e.g. assistant of searching articles of fruits and vegetables within over 40 classes.
Technologies: python, keras, cloud vision automl, gcp, fastapi, kubernetes.
Industry: retail

Recommender of items bought together

System discovers association rules by applying statistical analysis of probability of items to be purchased separately and conditional probability of items bought together. In order to explore the most relevant associations, it exploits apriory algorithm which is based on the following characteristics support, lift and confidence for each group of items.
Technologies: python, apriory, gcp.
Industry: retail

Fraud prevention

System collects behavioral characteristics during shopping trip of customer and predicts probability of potential fraud regarding to products scanned by customer.
Technologies: python, pythorch, gcp
Industry: retail

Customer retention

Model is trained on customers purchases history particularly it considers Recency, Frequency, and Monetary metrics to segment customers into homogeneous groups. The discovered patterns of customers behavior detects need of engagement with relevant campaigns to prevent customer churn.
Technologies: python, xgboost, bigquery, dataflow.
Industry: online content

Detection of violence of reviewing licensed online content

System analyzes images of video stream in real time to detect violence of license to review content by only person. In case of two or more people appear in the image taken from web camera or any attempt to record the content by phone is detected, then the message with corresponding image gets sent to administrator to inform about the suspicious behavior. In order to avoid latency issues, the model has been converted to run directly in browser.
Technologies: python, keras, tensorflow.js, django, django restframework.
Industry:L10n

Morphology service

System provides complex morphology analysis of given text data – part of speech tagging, lemmatization, tokenization (including Chinese, Korean and Japanese languages), extracting different types of named entities, glossary terms. System uses spacy models that are trained on custom data.
Technologies: python, spacy, flask, docker, AWS, CI/CD
Industry:L10n

Semantic Textual Similarity

Service helps clients to reduce localization costs by finding and grouping semantically similar strings in their projects before they go for translation.
Technologies: python, tensorflow, tf-serving, flask, docker, AWS, CI/CD
Industry:L10n

Aligning existing translations with proper source strings

Service helps to upload already translated files without identifier (html-like files, raw texts, etc) into localization management platform. It uses multilingual model to embed both source and translated strings into the same vector space and then finds the best possible match considering cosine distance, the file structure, and some other factors.
Technologies: python, tensorflow, tf-serving, flask, docker, AWS, CI/CD
Industry: e-commerce

Estimation of Sales and Share online retailer

Platform provides sales and market share estimates for clients’ products and competitors on one of large online retailers.
Technologies: python, pandas, sql, xgboost
Industry: e-commerce

Recommendation of target advertisement

The pipeline to determine the most probable candidates interesting in online education being based on information collected for site visitors.
Technologies: python, pandas, google maps api

Contacts

1A Sheptytskogo str. Ternopil, 46000, Ukraine
+38 096 771 85 80
suport@smartinsight.me

© 2021 smartinsight, Inc